Skip to content
Projects

chmod777john/swarm-ide has 1,496 GitHub stars and 174 forks. GitHot measured -1 stars in 24 hours. Last checked 2026-11-08 04:34 UTC. Last push 2026-03-02.

How these counts are measured

为 Agent Swarm 打造的 IDE ,支持 Kimi-2.5, GLM-4.7 等(即使不像 K2.5 那样经过强化学习的模型也能用)An IDE built for Agent Swarms, supporting Kimi-2.5, GLM-4.7, and more (compatible with models even without RL-reinforcement like K2.5).

Measured star growth

in 24h
-1
in 7 days
Update delayed
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

README_EN.md

Original

Swarm-IDE: Self-Organizing Agent Swarm

Demo

🎬 Video (Open‑source Kimi‑K2.5 Multi‑Agent Swarm)

WeChat Group

WeChat QR

Advantages

  • Create sub‑agents dynamically
  • Message any agent directly
  • WeChat‑style chat UI: intervene at any depth anytime
  • Streaming graph that visualizes collaboration in real time

Comparison

It is worth noting that this project independently introduced swarm mode before both Kimi-Swarm and Claude Team. Claude Team in particular shows striking overlap: with careful comparison, its core ideas (dynamic delegation and human-to-any-agent communication) align closely with this project’s design. To some extent, this reflects the advanced perspective behind the design. Building this independently at a time when static LangGraph workflows were dominant was notably ahead of its time. The whitepaper was timestamped on-chain back then; if you want to verify this directly, see the on-chain timestamp.

Item Kimi-Swarm Claude Agent Team Swarm-IDE
Nested agents ❌ ❌ ✅
Inter-agent messaging ❌ ✅ ✅
Human to sub-agent messaging ❌ ✅ ✅
Group chat mode ❌ ❌ ✅
Visualization ❌ ❌ ✅
Open source ❌ ❌ ✅
Release date 2026.1.27 2026.2.6 2026.1.2

UI Design

  • Graph shows swarm topology and live communication links
  • Tree‑structured multi‑level dialogs: talk to any agent like a chat app (even deep ones)
  • LLM history panel: shows the agent’s context; no more black box
  • Streaming tool‑call parameters in real time

Philosophy

  • Minimal primitives: only a few communication primitives are required (core is create + send)
  • Liquid topology: topology evolves during runtime; agents “hire” sub‑agents on demand
  • Flat collaboration: humans can step into any layer at any time

Concept

No complex nodes/edges abstraction — just think of “many people”:

Everyone can spawn children and talk to anyone.

With only these two capabilities, any structure can be built.

How to Run

cd agent-wechat
cd backend

cp .env.example .env.local
# Fill your keys and model names in .env.local

docker compose up -d
curl -X POST http://127.0.0.1:3017/api/admin/init-db
bun install
bun dev

Open http://localhost:3017

Click init-db, then create a workspace to start.

Try: “Create 3 children, ask them to message each other, then have each create 3 grandchildren.”

Zhihu

https://zhuanlan.zhihu.com/p/2000736341479138182

Environment Variables

The backend reads backend/.env.local:

  • OPENROUTER_API_KEY / OPENROUTER_MODEL (I use OpenRouter Kimi 2.5)
  • See backend/.env.example for all fields

Explore related projects